# The Capability Frontier: Benchmarks Miss 82% of Model Performance

arXiv: https://arxiv.org/abs/2606.26836 · HTML: https://mlsanity.com/paper/2606.26836/

## Key facts

- Cited by 1 papers in the map (data through 14 Oct 2026).
- Builds on 19 of its 38 references that are in the map.
- First posted on arXiv 25 Jun 2026.

## Builds on (most cited first, 19 of 19)

- [GPT-4 Technical Report](https://mlsanity.com/paper/2303.08774/) (2023)
- [Program Synthesis with Large Language Models](https://mlsanity.com/paper/2108.07732/) (2021)
- [TruthfulQA: Measuring How Models Mimic Human Falsehoods](https://mlsanity.com/paper/2109.07958/) (2021)
- [GPQA: A Graduate-Level Google-Proof Q&A Benchmark](https://mlsanity.com/paper/2311.12022/) (2023)
- [LiveCodeBench: Holistic and Contamination Free Evaluation of Large Language Models for Code](https://mlsanity.com/paper/2403.07974/) (2024)
- [BigCodeBench: Benchmarking Code Generation with Diverse Function Calls and Complex Instructions](https://mlsanity.com/paper/2406.15877/) (2024)
- [RouterBench: A Benchmark for Multi-LLM Routing System](https://mlsanity.com/paper/2403.12031/) (2024)
- [Large Language Model Routing with Benchmark Datasets](https://mlsanity.com/paper/2309.15789/) (2023)
- [Universal Model Routing for Efficient LLM Inference](https://mlsanity.com/paper/2502.08773/) (2025)
- [A Unified Approach to Routing and Cascading for LLMs](https://mlsanity.com/paper/2410.10347/) (2024)
- [MedCalc-Bench: Evaluating Large Language Models for Medical Calculations](https://mlsanity.com/paper/2406.12036/) (2024)
- [BEST-Route: Adaptive LLM Routing with Test-Time Optimal Compute](https://mlsanity.com/paper/2506.22716/) (2025)
- [Agreement-Based Cascading for Efficient Inference](https://mlsanity.com/paper/2407.02348/) (2024)
- [Cost-Aware Contrastive Routing for LLMs](https://mlsanity.com/paper/2508.12491/) (2025)
- [Efficient Training-Free Online Routing for High-Volume Multi-LLM Serving](https://mlsanity.com/paper/2509.02718/) (2025)
- [C3PO: Optimized Large Language Model Cascades with Probabilistic Cost Constraints for Reasoning](https://mlsanity.com/paper/2511.07396/) (2025)
- [LLMRank: Understanding LLM Strengths for Model Routing](https://mlsanity.com/paper/2510.01234/) (2025)
- [Breaking Model Lock-in: Cost-Efficient Zero-Shot LLM Routing via a Universal Latent Space](https://mlsanity.com/paper/2601.06220/) (2026)
- [Beyond Monoliths: Expert Orchestration for More Capable, Democratic, and Safe Language Models](https://mlsanity.com/paper/2506.00051/) (2025)

## Built on by (most cited first, 1 of 1)

- [What Does a Routing Oracle Measure Under Stochastic Decoding? Coupling, Scorer Choice, and Single-Commit Ceilings](https://mlsanity.com/paper/2607.03436/) (2026)

## Abstract (arXiv)

Existing benchmarks typically report accuracy for a single model on a single run. This systematically understates real-world LLM capabilities, particularly under heterogeneous data distributions: (i) different models get different questions correct according to their specializations, and (ii) given a budget, multiple generations can be sampled and selectively retained. To quantify this gap, we introduce the Capability Frontier: a Pareto frontier over a set of models that characterizes the best achievable performance at each cost level under optimal selection across models and generations (i.e., via an oracle). Our construction corrects for two opposing biases: underestimation from single-model evaluation and overestimation from taking maxima over noisy samples. We study 21 LLMs across 16 widely used benchmarks spanning coding, reasoning, medicine, factuality, instruction following, and agentic tasks, comparing Capability Frontier performance at matched cost to each benchmark's top-performing model. Correcting for single-model evaluation yields a 54% error rate reduction; additionally correcting for single runs yields an 82% improvement, with SOTA accuracy matched at 85% cost reduction. Complementing these empirical results, we use controlled probabilistic simulations to show that higher query topic entropy produces a near-monotonic increase in the performance gap between oracle routing and the best single model. Our findings suggest collective LLM capabilities are substantially underestimated, with implications for evaluation and deployment in data-heterogeneous, multi-domain settings.